Conf42 Platform Engineering 2026

Solanica’s Founder, Sergey Pronin, talks about Kubernetes, LLM and Platform Engineer walk into a bar.

Talk description

A platform engineer, an LLM, and Kubernetes walk into a bar. The platform engineer wants reproducibility. The LLM wants 80GB of VRAM and a GPU that isn’t already taken. Kubernetes wants everything to be a declarative object with a reconciler. Nobody gets what they want — at least not on the first try.

This talk is about what it actually takes to build Model-as-a-Service on top of OpenEverest: serving models with KServe and vLLM, routing through a unified AI gateway with per-model keys and token quotas, splitting models across multiple nodes when they don’t fit on one GPU, and autoscaling inference based on real traffic. We’ll show the architecture, the failure modes we hit, and the parts that turned out to be harder than expected.

If you’re a platform engineer who’s been handed “we need to run LLMs on our cluster” and told to figure it out, this is the talk we wish we’d had.

Watch the talk on Conf42